The weekly SecurityWeek overview of cybersecurity news provides a concise overview of important developments that may not receive comprehensive, stand-alone coverage, but remain relevant in the broader threat landscape.
This curated summary highlights top stories regarding vulnerability disclosures, emerging attack methods, policy updates, industry reports, and other noteworthy events to help readers stay well-informed about the evolving cybersecurity environment.
Here are this week’s highlights:
Anthropic Adds Security Center and Admin Controls to Claude Enterprise
Anthropic has deployed additional security and administration capabilities for Claude Enterprise, including a Security Center (in beta) that checks an organization’s Claude configuration for chat, Cowork, and Claude Code against Anthropic’s recommended settings. Inference hooks, now generally available, route prompts and tool calls to an organization’s existing data loss prevention and security tools for inspection, while a centralized console (also in beta) lets companies manage single sign-on, SCIM, and other settings across all of their Claude organizations from a single place.
Fake OpenAI and SpaceX event invites phishing of Microsoft and Google accounts
Push Security has documented a phishing campaign built around fake registration pages for OpenAI, SpaceX and Anthropic events, which dangle rewards such as three months of ChatGPT Pro or Grok Heavy to entice marketing employees to log in with their work accounts. The Microsoft option launches a device code phishing stream executed through the Kali365 phishing platform as a service. Push links the operator, with medium to high trust, to past fake Zoom event pages and says the campaign relies on Google and Zapier short links, Cloudflare challenges, and domains swapped every few weeks to avoid detection.
Chainalysis breach also exposed password hashes and tokens
Chainalysis expanded its breach disclosure, claiming to have discovered on November 2 that the names, usernames, password hashes and live tokens of all of its users had been exposed, after initially reporting that only non-sensitive data had been accessed. The company has since taken several products offline, including its KYT transaction monitoring service, after detecting abnormal activity, and is contacting customers whose KYT accounts show signs of unauthorized access.
Google entrusts Agent Substrate and gVisor to CNCF for AI agent sandboxing
Google is make a donation Agent Substrate, an open source orchestration layer on top of Kubernetes for running large numbers of isolated AI agent sessions, and gVisor, its open source application core, to the Cloud Native Computing Foundation (CNCF) as Sandbox projects. When an agent becomes inactive, Agent Substrate saves its state to storage and can restore it in less than 500 milliseconds. gVisor handles system calls in user space so that untrusted AI-generated code does not interact directly with the host kernel. According to Google, bringing both projects under the CNCF gives the community a shared, vendor-agnostic foundation for an agent infrastructure that can run on any Kubernetes cluster that meets compliance standards.
“Hello” messages hide Chinese-language attack framework targeting Flowise
Era researchers detailed To welcomea Chinese-language exploit framework that bombarded Flowise honeypots with thousands of chat messages reading only “hello” while abusing an old code injection flaw (CVE-2024-58351) to execute commands. The framework mapped the host, queried cloud metadata services, looked for SSH keys and other credentials, tried to configure persistence, and tested whether it could run payloads from memory. Era counted 4,372 Flowise instances exposed online, and of those whose version it was able to identify, 703 lacked authentication and 46 appeared to share the vulnerable configuration used in its honeypots.
LLM Blenders Turn Hacked Access to Enterprise AI into a Clandestine Activity
Mandiant says it investigated several cases of LLMJacking This year, attackers took control of companies’ access to AI models, generating fraudulent invoices of more than $100,000 per day and up to $1 million in total in some organizations. Initial access came primarily from stolen or exposed cloud credentials and API keys, and during an intrusion, attackers exploited a LiteLLM vulnerability (CVE-2026-40217) to upgrade ordinary keys to full administrator credentials. Google Threat Intelligence Group also found hacked access being sold through LLM mixers, including token routing service Jet Night, which it says uses stolen keys to offer unlimited plans with claimed frontier model access for around $60 per month.
Trezor sold by CryptoBilis would have been equipped with a seed-stealing implant
Trezor is asking customers who purchased its hardware wallets from CryptoBilis since March 2026 to transfer their funds to a freshly created wallet with a new seed, after a buyer reported it. find an implant inside a Trezor Safe 5 purchased from the former Southeast Asian dealer, who also sold Ledger devices that were found to be tampered with. According to the report, which Trezor has not yet verified, an additional circuit board equipped with a cellular module and an antenna captures the recovery phrase displayed on the screen and transmits it over a mobile network each time the device is connected. No stolen funds have been reported.
Harvard alumni and donor data from 2025 breach added to HIBP
Have I been pwned uploaded approximately 851,000 unique email addresses from the November 2025 Harvard University breach, an incident later linked to an extortion bid by ShinyHunters. The records come from Harvard’s alumni and fundraising systems and contain detailed profiles of donors, alumni and parents, as well as certain faculty, staff and students, including names, contact information, dates and places of birth, employers, ethnicities, marital status and donation history.
SAP’s operating window shrinks to days as interest in ransomware grows
Onapsis and recorded future report that for SAP vulnerabilities patched in 2025 and 2026, the median gap between the release of a security note and the first verified exploit or public proof of concept was one to three days, compared to 489 days for vulnerabilities patched in 2021. Ransomware and extortion groups appearing in SAP-related content increased from 8 in 2023 to 26 in 2025, and SAP CVEs with confirmation wild exploitation or public proof of concept code doubled from 9 to 18 between late 2023 and mid-2026. The study, which draws on more than 1.2 million threat intelligence events, also tracked clandestine SAP access lists that range in price from $500 to $30,000.
Synology flaw patched in 2024 now exploited in ransomware attacks
Synology warns that attackers are exploiting a vulnerability patched in December 2024 to strike devices still running outdated DSM versions, with the aim of encrypting data and extorting victims. The company did not identify the specific defect, only pointing out its Synology-SA-24:20 advisory. Synology recommends that anyone using a vulnerable version update DSM immediately and enable automatic updates.
Alleged Nomad Bridge operator appears in US court after extradition
Alexander Gurevich, a 49-year-old Israeli national now known as Alexander Block, appeared in a US court for the first time on October 30, a day after Israel extradited him to face charges that he exploited the Nomad Bridge smart contract in August 2022 to obtain virtual currency worth approximately $2.89 million. According to the indictment, he exchanged the stolen WBTC and WETH for other cryptocurrencies and spread them across multiple addresses to disguise their origin. The money laundering charge alone carries a maximum sentence of 20 years in prison.
Related: In Other news: Tor output relays hijacked, banks spot malicious AI agents, self-rebuilding macOS ransomware
Related: In Other news: Dispute over an anthropogenic violation, UEFI flaw affects most servers, Cloudflare AI data leak
